Application and evaluation of Machine Learning for news article popularity prediction

Abstract

The internet is increasingly becoming the primary source of news worldwide. Social networking sites have further enabled instantaneous spread of such articles by often allowing single-click user sharing. Majority of the organizations publishing such articles drive revenue through advertisements which is ultimately dependent on the popularity of the article. This popularity is mainly defined in terms of views and shares. One of the emerging applications of Machine Learning is to help organizations predict which articles are most likely to become popular and thus allow them to improve targeted advertising campaigns in order to optimize revenue. This paper proposes and evaluates Machine Learning based approaches alongside Rolling, Growing and a Hybrid training window techniques in order to predict the popularity of news articles.